!49121 Support None is input of real kernel.

Merge pull request !49121 from Margaret_wangrui/new_fallback_none_real_kernel

mindspore/ccsrc/frontend/optimizer/clean.cc

@@ -34,6 +34,7 @@
 #include "ir/value.h"
 #include "pipeline/jit/parse/resolve.h"
 #include "utils/hash_map.h"
+#include "utils/anf_utils.h"
 
 namespace mindspore {
 /* namespace to support opt */
@@ -969,6 +970,9 @@ class CleanAfterOptARewriter : public BaseRewriter {
     if (!support_fallback_runtime) {
       return;
     }
+    if (AnfUtils::IsRealKernel(cnode)) {
+      return;
+    }
     const auto &inputs = cnode->inputs();
     const auto &cur_func = cnode->func_graph();
     MS_EXCEPTION_IF_NULL(cur_func);

mindspore/ccsrc/frontend/optimizer/irpass/convert.h

@@ -70,8 +70,7 @@ class PrintConstStringWrapper : public AnfVisitor {
       return std::any_of(elements.cbegin(), elements.cend(),
                          [&](const AbstractBasePtr &ele) { return CheckNeedConvert(ele); });
     }
-    return !abs->isa<abstract::AbstractScalar>() && !abs->isa<abstract::AbstractTensor>() &&
-           !abs->isa<abstract::AbstractNone>();
+    return !abs->isa<abstract::AbstractScalar>() && !abs->isa<abstract::AbstractTensor>();
   }
 
   AnfNodePtr ConvertString(const AbstractBasePtr &abs) const {

mindspore/ccsrc/pybind_api/ir/py_execute_py.h

@@ -69,7 +69,6 @@ class PyExecuteInitializer {
     // so special handling of None is required.
     if (script->ToString() == "None") {
       const auto &output = py::none();
-      MS_LOG(DEBUG) << "Python output type: " << py::str(output.get_type()) << ", output: " << output;
       PushPyExecuteOutput(output);
       const auto &infer_shape = std::make_shared<abstract::Shape>(ShapeVector({1}));
       return abstract::MakeAbstract(infer_shape, kFloat64);

mindspore/ccsrc/utils/convert_utils_py.cc

@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2021 Huawei Technologies Co., Ltd
+ * Copyright 2019-2023 Huawei Technologies Co., Ltd
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -493,19 +493,14 @@ py::object VectorRefToPyData(const VectorRef &value_list, const AbstractBasePtr
     }
     return ref_tuple;
   }
-  // The size of seq_abs may be larger than the size of value_list, because the backend will eliminate None.
   size_t ref_idx = 0;
   for (size_t i = 0; i < seq_abs->size(); i++) {
     auto elem_abs = seq_abs->elements()[i];
-    if (elem_abs->isa<abstract::AbstractNone>()) {
-      continue;
-    }
     ref_tuple[ref_idx] = BaseRefToPyData(value_list[ref_idx], elem_abs);
     ref_idx++;
   }
   if (ref_idx != value_size) {
-    MS_LOG(EXCEPTION) << "The size of elements (excluding None) should be equal to " << value_size << ", but got "
-                      << ref_idx;
+    MS_LOG(EXCEPTION) << "The size of elements should be equal to " << value_size << ", but got " << ref_idx;
   }
   ret = ref_tuple;
   return ret;

tests/st/fallback/test_graph_fallback_none.py

@@ -16,8 +16,11 @@ import pytest
 import numpy as np
 
 from mindspore import Tensor, jit, context, Parameter
+from mindspore.nn import Cell
+from mindspore.nn.probability import distribution
 import mindspore.nn as nn
 import mindspore.common.dtype as mstype
+import mindspore as ms
 
 context.set_context(mode=context.GRAPH_MODE)
 
@@ -222,7 +225,6 @@ def test_none_is_slice_in_list():
     assert res == 0
 
 
-@pytest.mark.skip(reason="No support print None.")
 @pytest.mark.level0
 @pytest.mark.platform_x86_gpu_training
 @pytest.mark.platform_arm_ascend_training
@@ -397,3 +399,44 @@ def test_none_is_input_of_tuple_return():
 
     out = foo()
     assert out == (1, "a", None)
+
+
+@pytest.mark.level0
+@pytest.mark.platform_x86_gpu_training
+@pytest.mark.platform_arm_ascend_training
+@pytest.mark.platform_x86_ascend_training
+@pytest.mark.env_onecard
+def test_none_is_input_of_tuple_return_2():
+    """
+    Feature: Support None.
+    Description: Support None is input of tuple, and the tuple is return.
+    Expectation: No exception.
+    """
+    class BernoulliCrossEntropy(Cell):
+        def __init__(self, probs, seed=10, dtype=ms.int32, name='Bernoulli', dist='Bernoulli'):
+            super().__init__()
+            self.b = distribution.Bernoulli(probs, seed, dtype, name)
+            self.dist = dist
+
+        def construct(self, probs1_b, probs1=None):
+            if probs1 is None:
+                out1 = self.b.cross_entropy(self.dist, probs1_b)
+                out2 = self.b.kl_loss(self.dist, probs1_b)
+            else:
+                out1 = self.b.cross_entropy(self.dist, probs1_b, probs1)
+                out2 = self.b.kl_loss(self.dist, probs1_b, probs1)
+            out3 = self.b.probs
+            return out1, out2, out3
+
+    probs = None
+    probs1 = 0.2
+    probs1_b = 0.9
+    context.set_context(mode=context.GRAPH_MODE)
+    net_graph = BernoulliCrossEntropy(probs)
+    out_me_graph = net_graph(Tensor(probs1_b), Tensor(probs1))
+    print("out_me_graph: ", out_me_graph)
+    context.set_context(mode=context.PYNATIVE_MODE)
+    net_pynative = BernoulliCrossEntropy(probs)
+    out_me_pynative = net_pynative(Tensor(probs1_b), Tensor(probs1))
+    print("out_me_pynative: ", out_me_pynative)
+    assert out_me_graph == out_me_pynative

tests/st/ops/ascend/test_gru_op.py

@@ -13,7 +13,6 @@
 # limitations under the License.
 # ==============================================================================
 
-import os
 import math
 import pytest
 import numpy as np
@@ -94,7 +93,6 @@ class GRUWeightBias():
 @pytest.mark.platform_x86_ascend_training
 @pytest.mark.env_onecard
 def test_sit_gru_forward_input_3_32_32_is_32_hs_16():
-    os.environ['MS_DEV_ENABLE_FALLBACK_RUNTIME'] = '0'
     input_size = 32
     hidden_size = 16
     has_bias = True
@@ -117,7 +115,6 @@ def test_sit_gru_forward_input_3_32_32_is_32_hs_16():
     net.gru.b_ih_list = b_ih_list
     net.gru.b_hh_list = b_hh_list
     out, hy = net(input_ms, h0)
-    os.environ['MS_DEV_ENABLE_FALLBACK_RUNTIME'] = '1'
 
     # pynative mode
     context.set_context(mode=context.PYNATIVE_MODE)
@@ -138,7 +135,6 @@ def test_sit_gru_forward_input_3_32_32_is_32_hs_16():
 @pytest.mark.platform_x86_ascend_training
 @pytest.mark.env_onecard
 def test_sit_gru_grad_input_3_32_32_is_32_hs_16():
-    os.environ['MS_DEV_ENABLE_FALLBACK_RUNTIME'] = '0'
     input_size = 32
     hidden_size = 16
     has_bias = True
@@ -166,7 +162,6 @@ def test_sit_gru_grad_input_3_32_32_is_32_hs_16():
     out_grad, _ = grad_net_inp(input_ms, h0)
     x_grad = out_grad[0].asnumpy()
     h_grad = out_grad[1].asnumpy()
-    os.environ['MS_DEV_ENABLE_FALLBACK_RUNTIME'] = '1'
 
     # pynative mode
     context.set_context(mode=context.PYNATIVE_MODE)

tests/st/ops/ascend/test_lstm_op.py

@@ -13,7 +13,6 @@
 # limitations under the License.
 # ==============================================================================
 
-import os
 import math
 import pytest
 import numpy as np
@@ -94,7 +93,6 @@ class LSTMWeightBias():
 @pytest.mark.platform_x86_ascend_training
 @pytest.mark.env_onecard
 def test_sit_lstm_forward_input_3_32_32_is_32_hs_16():
-    os.environ['MS_DEV_ENABLE_FALLBACK_RUNTIME'] = '0'
     input_s = 32
     hidden_s = 16
     has_bias = True
@@ -118,7 +116,6 @@ def test_sit_lstm_forward_input_3_32_32_is_32_hs_16():
     net.lstm.b_ih_list = b_ih_list
     net.lstm.b_hh_list = b_hh_list
     out, (hy, cy) = net(input_ms, h0, c0)
-    os.environ['MS_DEV_ENABLE_FALLBACK_RUNTIME'] = '1'
 
     # pynative mode
     context.set_context(mode=context.PYNATIVE_MODE)
@@ -141,7 +138,6 @@ def test_sit_lstm_forward_input_3_32_32_is_32_hs_16():
 @pytest.mark.platform_x86_ascend_training
 @pytest.mark.env_onecard
 def test_sit_lstm_grad_input_3_32_32_is_32_hs_16():
-    os.environ['MS_DEV_ENABLE_FALLBACK_RUNTIME'] = '0'
     input_s = 32
     hidden_s = 16
     has_bias = True
@@ -171,7 +167,6 @@ def test_sit_lstm_grad_input_3_32_32_is_32_hs_16():
     x_grad = out_grad[0].asnumpy()
     h_grad = out_grad[1].asnumpy()
     c_grad = out_grad[2].asnumpy()
-    os.environ['MS_DEV_ENABLE_FALLBACK_RUNTIME'] = '1'
 
     # pynative mode
     context.set_context(mode=context.PYNATIVE_MODE)